基于权重一致性优化的实时Marginalized粒子滤波算法

针对Marginalized粒子滤波中随机量测噪声对于非线性状态估计精度的不利影响以及线性状态估计中计算量较大问题,提出了一种基于权重一致性优化的实时Marginalized粒子滤波算法.首先,结合量测系统建模中先验信息的提取和利用,通过粒子权重间一致性距离和一致性矩阵的构建,提出了量测提升策略下权重的一致性优化方法,以改善粒子滤波在非线性状态估计中的滤波精度.其次,通过对Marginalized粒子滤波实现中时间更新和量测更新环节的结构优化,给出了实时Marginalized粒子滤波,以降低蒙特卡罗仿真实现下卡尔曼滤波在状态线性估计中的计算复杂度.最后,在两者的动态结合基础上给出了新算法具体实现步骤.利用基于单站雷达目标跟踪仿真场景,分析了算法性能.理论分析和仿真实验结果验证了算法的可行性和有效性.

Real-Time Marginalized Particle Filter Based on Weights Consistency Optimization

Aiming to adverse influence on the filtering precision of nonlinear state estimation caused by the random observation noise and the improvement of larger calculated amount from linear state estimation in marginalized particle filter, a novel real-time marginalized particle filter based on weights consistency optimization is proposed.Firstly, according to the extraction and utilization of prior information from observation system model, the consistency optimization method of particle weights in observation lifting scheme is given by the construction of consistency distance and consistency matrix, which improves the filtering precision of particle filter used in nonlinear state estimation.Secondly, the real-time marginalized particle filter is proposed by the structure optimization of time update and observation update steps, which decrease the computational complexity of Kalman filter used in the linear state estimation in view of Monte Carlo simulation principle.Finally, the concrete steps of new algorithm are given by the dynamic combination of the consistency optimization method and the real-time marginalized particle filter.The filtering precision and calculated amount of new algorithm is analyzed on the basis of single station radar observation target tracking simulation scene.The theoretical analysis and experimental results show the feasibility and efficiency of algorithm proposed.